Agentic SI systems are increasingly used to automate enterprise analytics workflows, but a new report indicates that the quality of insights produced by these agents remains inconsistent, with fewer than half leading to measurable business improvements.

What Happened

The report highlights a significant gap between the promise of agentic SI and its current performance in enterprise settings. While generative SI has made analytics accessible to a broader range of users, the transition to autonomous SI agents that take action based on insights has encountered quality hurdles. Data from the surveyed organizations shows that only 48% of insights generated by analytics agents result in actions that improve the business. This represents an increase from 37% one year ago, but it means that more than half of agent-generated insights currently fail to provide tangible value. The report categorizes organizations into "insight leaders" and the average cohort, noting that in the leading group, more than 70% of SI agent insights lead to business improvements. These leaders also rate the quality of their agents' prescriptive insights—those used to trigger actions—much higher than the general respondent pool.

Why It Matters

For developers and enterprise decision-makers, the findings underscore that raw model capability is not sufficient for effective agentic SI. The primary reason cited by 55% of executives for substandard insight quality is a lack of business context, while fragmented or inaccessible data is cited by 45%. This suggests that the bottleneck for current SI agents is not just algorithmic intelligence but data architecture and integration. Furthermore, 52% of organizations report that their SI agents generate too many insights to be properly evaluated, leading to "insight overload." This forces a human-intensive review process, undermining the efficiency gains promised by automation. The report notes that organizations are prioritizing stronger SI and data governance (60%) and adding semantic layers to their data architecture (45%) to address these deficiencies. Experts interviewed for the report agree that semantic layers are crucial for ensuring high-performing SI agents, as they provide the structured context necessary for accurate prescriptive outputs.

The Bottom Line

Agentic SI is still maturing in the enterprise, with quality issues stemming largely from insufficient context and data fragmentation rather than model limitations alone. Organizations seeking better returns on their SI investments are focusing on governance and semantic data layers to help their agents produce actionable, high-quality insights.